From: Hyonseok Hwang (danggi_at_northwestern.edu)
Date: Tue Mar 02 2004 - 13:18:22 CST
Thank you very much, Jim. Your answer is very helpful. Now I
understand them more clearly.
Could you let me know any references on Langevin temperature control
method used in NAMD, please?
Jérôme Hénin wrote:
>>just to re-ask some questions regarding temperature control which I
>>don't think have been answered (Satyavani Vemparala 25/2/04, Hyonseok
>>Hwang 24/2/04, Wei Fu 20/2/04, Hyonseok Hwang 9/2/04).
>>How does one decide what value to use for the langevin coupling
>>coefficient (0.1, 5, 10?), or what is an appropriate value?
> The Langevin damping coefficient determines how fast the atoms "forget" their
> momentum, and how fast the lost energy is re-introduced by stochastic forces.
> The bigger the coefficient, the faster temperature fluctuations are
> compensated, and the farther the dynamics is from Newtonian motion (and
> closer to a purely stochastic motion).
> If you're interested in dynamic properties, you'll want to use coefficients as
> small as possible (while still keeping the temperature reasonably constant).
> For the biological systems I've simulated, values around 1 ps-1 were good
> trade-offs. You should only need more if strongly exothermal phenomena happen
> in your system - maybe when starting the equilibration of a new system - but
> then usually you don't mind too much if temperature raises a little for a
> short time.
>>Is there any relationship between the langevin coupling coefficient and
>>the temperature coupling coefficient in Berendsen's method?
> I believe this question was not answered previously because it is not easy to
> see what kind of relationship you are looking for. Since the algorithms are
> really different, what relationship could there be between their parameters ?
> Both coefficients are the reciprocal of a characteristic time of the
> thermostat, and both are related to a friction coefficient, but as far as I
> can see, the similarity ends there. For example, the Berendsen friction
> coefficient depends on time, whereas the one in Langevin does not.
> Also I'm not sure I understand what the use of such a relationship would be.
> If you want a Berendsen coupling simulation and a Langevin simulation to be
> equivalent, don't forget that Berendsen coupled MD trajectories do not sample
> from the canonical ensemble.
> Maybe someone on the list will have something to add on these points, though.
This archive was generated by hypermail 2.1.6 : Wed Feb 29 2012 - 15:38:28 CST